[
https://issues.apache.org/jira/browse/CASSANDRA-19776?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17950873#comment-17950873
]
Stefan Miklosovic edited comment on CASSANDRA-19776 at 5/12/25 8:55 AM:
------------------------------------------------------------------------
To repeat what we saw, for other people to jump in more easily, as it might be
quite chaotic to understand the symptoms:
1) a compaction is happening
2) somebody calls above-mentioned JMX metric
3) that method spins / is stuck, because CANONICAL returns all tables, even
expired, but there is no reference to that
4) compaction does not take into account expired sstables when doing "try
(Refs<SSTableReader> refs = Refs.ref(actuallyCompact);" (actuallyCompact will
_not_ contain expired ones), expired ones are just a logical part of
compaction, but I do not see that we would actually reference them
5) because we have not referenced expired ones in 4), there is nobody
referencing them, so selectAndReference invoked via metric spins
Once compaction strategy finishes / when expired SSTables are removed, then
CANONICAL will not contain any expired SSTables and the metric method ends
relatively fast.
That is my understanding of that.
What we try to do is that we would grab references to expired SSTables as well
while compacting, just so selectAndReference on CANONICAL has something to
select and reference in order to not spin, but then the question is if it is
problematic to reference expired in compaction like that ....
was (Author: smiklosovic):
To repeat what we saw, for other people to jump in more easily, as it might be
quite chaotic to understand the symptoms:
1) a compaction is happening
2) somebody calls above-mentioned JMX metric
3) that method spins / is stuck, because CANONICAL returns all tables, even
expired, but there is no reference to that
4) compaction does not take into account expired sstables when doing "try
(Refs<SSTableReader> refs = Refs.ref(actuallyCompact);" (actuallyCompact will
_not_ contain expired ones), expired ones are just a logical part of
compaction, but I do not see that we would actually reference them
5) because we have not referenced expired ones in 4), there is nobody
referencing them, so selectAndReference
Once compaction strategy finishes / when expired SSTables are removed, then
CANONICAL will not contain any expired SSTables and the metric method ends
relatively fast.
That is my understanding of that.
What we try to do is that we would grab references to expired SSTables as well
while compacting, just so selectAndReference on CANONICAL has something to
select and reference in order to not spin, but then the question is if it is
problematic to reference expired in compaction like that ....
> Spinning trying to capture readers
> ----------------------------------
>
> Key: CASSANDRA-19776
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19776
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Legacy/Core
> Reporter: Cameron Zemek
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 4.0.x, 4.1.x, 5.0.x, 5.x
>
> Attachments: extract.log
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> On a handful of clusters we are noticing Spin locks occurring. I traced back
> all the calls to the EstimatedPartitionCount metric (eg.
> org.apache.cassandra.metrics:type=Table,keyspace=testks,scope=testcf,name=EstimatedPartitionCount)
> Using the following patched function:
> {code:java}
> public RefViewFragment selectAndReference(Function<View,
> Iterable<SSTableReader>> filter)
> {
> long failingSince = -1L;
> boolean first = true;
> while (true)
> {
> ViewFragment view = select(filter);
> Refs<SSTableReader> refs = Refs.tryRef(view.sstables);
> if (refs != null)
> return new RefViewFragment(view.sstables, view.memtables,
> refs);
> if (failingSince <= 0)
> {
> failingSince = System.nanoTime();
> }
> else if (System.nanoTime() - failingSince >
> TimeUnit.MILLISECONDS.toNanos(100))
> {
> List<SSTableReader> released = new ArrayList<>();
> for (SSTableReader reader : view.sstables)
> if (reader.selfRef().globalCount() == 0)
> released.add(reader);
> NoSpamLogger.log(logger, NoSpamLogger.Level.WARN, 1,
> TimeUnit.SECONDS,
> "Spinning trying to capture readers {},
> released: {}, ", view.sstables, released);
> if (first)
> {
> first = false;
> try {
> throw new RuntimeException("Spinning trying to
> capture readers");
> } catch (Exception e) {
> logger.warn("Spin lock stacktrace", e);
> }
> }
> failingSince = System.nanoTime();
> }
> }
> }
> {code}
> Digging into this code I found it will fail if any of the sstables are in
> released state (ie. reader.selfRef().globalCount() == 0).
> See the extract.log for an example of one of these spin lock occurrences.
> Sometimes these spin locks last over 5 minutes. Across the worst cluster with
> this issue, I ran a log processing script that everytime the 'Spinning trying
> to capture readers' was different to previous one it would output if the
> released tables were in Compacting state. Every single occurrence has it spin
> locking with released listing a sstable that is compacting.
> In the extract.log example its spin locking saying that nb-320533-big-Data.db
> has been released. But you can see prior to it spinning that sstable is
> involved in a compaction. The compaction completes at 01:03:36 and the
> spinning stops. nb-320533-big-Data.db is deleted at 01:03:49 along with the
> other 9 sstables involved in the compaction.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
